Amalgamating Filtered Knowledge: Learning Task-customized Student from Multi-task Teachers

TL;DR

This paper proposes a method for training a student CNN to learn task-specific knowledge from multiple pre-trained teacher networks without human annotations, resulting in a versatile model that can outperform its teachers on customized tasks.

Contribution

It introduces a layer-wise training strategy to effectively amalgamate knowledge from multiple teachers for task-specific student learning without labeled data.

Findings

Student models outperform teachers on customized tasks.

Layer-wise training effectively filters relevant knowledge.

Method works across multiple benchmarks.

Abstract

Many well-trained Convolutional Neural Network(CNN) models have now been released online by developers for the sake of effortless reproducing. In this paper, we treat such pre-trained networks as teachers and explore how to learn a target student network for customized tasks, using multiple teachers that handle different tasks. We assume no human-labelled annotations are available, and each teacher model can be either single- or multi-task network, where the former is a degenerated case of the latter. The student model, depending on the customized tasks, learns the related knowledge filtered from the multiple teachers, and eventually masters the complete or a subset of expertise from all teachers. To this end, we adopt a layer-wise training strategy, which entangles the student's network block to be learned with the corresponding teachers. As demonstrated on several benchmarks, the learned student network achieves very promising results, even outperforming the teachers on the customized tasks.